Consumer Cost-Sharing in Primary Care: Unintended Health and Economic Outcomes

在许多国家，即使是有保险的个人也必须支付费用（通常是病人分担费用）才能看家庭医生。收费的理论目的是控制成本：限制过度看医生，尽管它也可以在为卫生系统提供资金方面发挥作用。这种收费的一个意外后果是，它们可能会阻止个人在必要的医疗条件下看家庭医生。因此，个人的健康可能会恶化，将来他们可能需要更昂贵的医疗（例如住院），这将使费用本来应该达到的控制成本的目的落空。随着非传染性疾病（NCD）的迅速增加，这种意外后果的重要性可能会增加，这需要通过初级保健服务进行及时诊断和管理。对于许多非传染性疾病，推迟看医生比较容易，因为它们在最初阶段并不痛苦（例如糖尿病），但如果得不到及时诊断和适当管理，它们将导致未来更昂贵的医疗程序。看家庭医生可能有助于及时诊断病情，并对这些病情进行适当的管理。因此，病人的费用可能有利于使用医院而不是基层医疗服务，这是效率低下的，因为医院服务要昂贵得多。这种低效率削弱了卫生系统，并限制了卫生系统在其他方面（覆盖率，质量改善）的改善程度。尽管文献对这一主题感兴趣，但大多数以前的研究报告了相关性，这可能是虚假的。最近的一些论文已经能够估计病人费用对健康的影响，但他们还没有能够评估医疗保健使用模式或整体治疗成本的变化。这些是了解患者费用如何影响卫生系统（初级和二级保健之间的资源分配）及其效率的关键问题。为了促进这一辩论，我们将测试是否初级保健中增加的患者费用（以及增加多少）会增加未确诊的慢性病、不良健康结果、死亡率、医院服务的使用以及短期和长期的治疗费用（最多7年）。为了开展这项工作，我们将使用2011年至2018年的卫生行政数据，覆盖97%的哥伦比亚人口，包含哥伦比亚卫生系统提供的所有卫生保健服务的患者记录，包括使用服务的日期和类型（门诊、医院等）、处方、治疗费用、ICD-10、个人的社会人口特征（包括收入或财富评分）和死亡率。个人识别码在七年中保持一致，提供了一个独特的丰富和详细的纵向行政数据库。此外，它的巨大规模使我们能够估计特定亚群的影响（例如，社会经济状况不佳的个人或慢性病患者）。然而，数据不足以为感兴趣的问题提供可靠的答案。我们还需要一种方法来确保我们不会在数据中报告虚假的关联。实验通常用于此目的，但它们不太可能像我们将要估计的那样为我们提供长期影响，样本也不够大。我们很幸运，哥伦比亚的患者成本分摊系统在“突然跳跃”中工作，也就是说，成本分摊在某些连续变量的预先指定的阈值上突然跳跃。这是应用称为回归不连续性（RD）的准实验方法的理想设置，该方法已知在非常弱的假设下提供因果估计，没有虚假相关性。请注意，你不能随时使用RD，条件必须在那里，但我们很幸运，他们在哥伦比亚举行。

In many countries, even insured individuals must pay a fee (usually patient cost-sharing) to see their family doctor. The theoretical purpose of that fee is one of cost-containment: to limit the overuse of doctor visits, although it can also play a role in the funding the health system.An unintended consequence of such fees is that they might prevent individuals from visiting their family doctor for necessary medical conditions. Hence, individuals' health might deteriorate, and in the future they might need much more expensive medical treatments (e.g. hospitalizations), which would defeat the cost containment purpose that the fee was supposed to serve. The importance of this unintended consequence might be growing with the rapid increase in Non-Communicable Diseases (NCDs), which require timely diagnosis and management through primary care services. For many NCDs, it is easier to postpone doctor visits because they are not painful in their initial stages (e.g. diabetes), but if they are not diagnosed timely and appropriately managed, they will lead to more expensive medical procedures in the future. Visiting the family doctor might help to diagnose the conditions timely, as well as to keep an adequate management of such conditions.Hence, patient fees might be favoring use of hospitals instead of primary care services, which is inefficient because hospital services are much costlier. This inefficiency weakens the health system and limits how much the health system can improve in other dimensions (coverage, quality improvement).Although the literature has been interested in this topic, most previous research has reported associations, which might be spurious. Some recent papers have been able to estimate the effect of patient fees on health, but they have not been able to assess how health care use patterns or overall treatment costs change. These are key issues to understand how patient fees affect the health system (split of resources between primary and secondary care), and its efficiency. To contribute to this debate, we will be testing whether (and by how much) increased patient fees in primary care increase undiagnosed chronic conditions, adverse health outcomes, mortality, use of hospital services, and treatment costs both in the short and long term (up to 7 years).To conduct this work, we will be using health administrative data for the years 2011 to 2018, covering 97% of the Colombian population and containing patients records of all health care services provided in the Colombian Health System, including date and type of service used (outpatient, hospital, etc), prescriptions, treatment costs, ICD-10, sociodemographic characteristics of individuals (including income or wealth scores) and mortality. The person identifier is consistent across the seven years, providing a uniquely rich and detailed longitudinal administrative database. Moreover, its huge size allows us to estimate the effects of interest for particular subpopulations of interest (e.g. individuals with poor socio-economic status, or chronic patients).However, data is not enough to provide a robust answer to the question of interest. We also need a method to be sure that we will not be reporting spurious associations in the data. Experiments are usually used for that purpose but they are unlikely to provide us with long term effects as the ones that we will be estimating, nor the samples be large enough.We are fortunate enough that the patient cost-sharing system in Colombia works "in abrupt jumps," that is, cost-sharing jumps abruptly at pre-specified thresholds of some continuous variables. This is the ideal setting to apply a quasi-experimental method called Regression Discontinuity (RD), which is known to provide causal estimates, free of spurious correlations, under very weak assumptions. Note that you cannot use RD whenever you want, the conditions must be there, but we are fortunate that they do hold in Colombia.